CN115277626B - Address information conversion method, electronic device, and computer-readable storage medium - Google Patents
Address information conversion method, electronic device, and computer-readable storage medium Download PDFInfo
- Publication number
- CN115277626B CN115277626B CN202210908906.8A CN202210908906A CN115277626B CN 115277626 B CN115277626 B CN 115277626B CN 202210908906 A CN202210908906 A CN 202210908906A CN 115277626 B CN115277626 B CN 115277626B
- Authority
- CN
- China
- Prior art keywords
- training
- address information
- recognition model
- address
- longitude
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L61/00—Network arrangements, protocols or services for addressing or naming
- H04L61/09—Mapping addresses
- H04L61/25—Mapping addresses of the same type
- H04L61/2503—Translation of Internet protocol [IP] addresses
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
- Y02D10/00—Energy efficient computing, e.g. low power processors, power management or thermal management
Landscapes
- Engineering & Computer Science (AREA)
- Computer Networks & Wireless Communication (AREA)
- Signal Processing (AREA)
- Character Discrimination (AREA)
Abstract
The present invention relates to the field of artificial intelligence, and in particular, to an address information conversion method, an electronic device, and a computer readable storage medium. In the address information conversion method of the embodiment of the invention, the target address information is firstly acquired, and the target address information is input into the Seq2Seq semantic recognition model. Further, semantic feature extraction is carried out on target address information through a Seq2Seq semantic recognition model to obtain address information features corresponding to the target address information, and then the Seq2Seq semantic recognition model is used for recognizing the address information features to obtain target longitude and latitude coordinates corresponding to the target address information. The address information conversion method of the embodiment of the invention converts the target address information into the target longitude and latitude coordinates based on the Seq2Seq semantic recognition model, does not need to rely on a network to perform online inquiry, and can realize the conversion of the address information into the corresponding longitude and latitude coordinates in an offline state.
Description
Technical Field
The present invention relates to the field of artificial intelligence, and in particular, to an address information conversion method, an electronic device, and a computer readable storage medium.
Background
Along with development of science and technology, in many occasions corresponding to location addresses and artificial intelligence services, address information needs to be converted into longitude and latitude, so that text semantics are converted into information needed by positioning. In the related art, the conversion of the address information into longitude and latitude is generally required, text semantics of the address information are firstly identified, then the address information is mapped into a longitude and latitude coordinate network according to administrative division such as province and city and county, and the like, and verification is carried out on the address information and the known address in an address database, so that the method is realized. However, the above method often requires accessing an address database on line, or building an extremely large address database in the local institution, so that it is difficult to convert address information into corresponding longitude and latitude in an offline situation. Therefore, how to convert address information into corresponding longitude and latitude coordinates in an offline state is a problem to be solved urgently by those skilled in the art.
Disclosure of Invention
The present invention aims to solve at least one of the technical problems existing in the prior art. To this end, the present invention proposes an address information conversion method, an electronic device, and a computer-readable storage medium, capable of implementing conversion of address information into corresponding longitude and latitude coordinates in an offline state.
An address information conversion method according to an embodiment of the first aspect of the present invention includes:
acquiring target address information, and inputting the target address information into a Seq2Seq semantic recognition model, wherein the Seq2Seq semantic recognition model is obtained by training a basic recognition model through training address information and training longitude and latitude coordinates corresponding to the training address information;
extracting semantic features of the target address information through the Seq2Seq semantic recognition model to obtain address information features corresponding to the target address information;
and identifying the address information features through the Seq2Seq semantic identification model to obtain the target longitude and latitude coordinates corresponding to the target address information.
According to some embodiments of the invention, before the obtaining the target address information and inputting the target address information into the Seq2Seq semantic recognition model, the method further includes:
acquiring a training data set, wherein the training data set comprises training address information and training longitude and latitude coordinates corresponding to the training address information in a target region range, and the target region range comprises a target place pointed by the target address information;
And carrying out optimization training on the basic recognition model based on the training data set to obtain the Seq2Seq semantic recognition model.
According to some embodiments of the invention, the optimizing training the basic recognition model based on the training data set to obtain the Seq2Seq semantic recognition model includes:
performing iterative training on the basic recognition model based on the training data set, and inputting the training address information into the basic recognition model for performing the training treatment of each round in each round of iterative training to acquire the recognition coordinates of each round corresponding to the training address information;
after each round of iterative training, comparing the identification coordinates of each round with the longitude and latitude coordinates of each training, calculating the identification accuracy of the basic identification model and updating the parameters of the basic identification model;
and counting the change condition of the recognition accuracy after each round of iterative training, and obtaining the trained Seq2Seq semantic recognition model when the recognition accuracy is converged to a first certain value.
According to some embodiments of the present invention, the performing iterative training on the basic recognition model based on the training data set, in each round of the iterative training, inputting the respective training address information into the basic recognition model to perform a present round of training processing, and obtaining respective present round recognition coordinates corresponding to the respective training address information, including:
In each round of iterative training, randomly starting a batch of semantic recognition neurons in the basic recognition model;
and carrying out the round of training processing on the training address information based on the semantic recognition neurons which are randomly opened, and obtaining the recognition coordinates of each round.
According to some embodiments of the invention, after each round of the iterative training, comparing the recognition coordinates of each round with the longitude and latitude coordinates of each training, calculating the recognition accuracy of the basic recognition model and updating the parameters of the basic recognition model, including:
obtaining each coordinate interval error according to the identification coordinates of each own wheel and the corresponding training longitude and latitude coordinates;
acquiring the number of accurate results of which the coordinate interval error is smaller than a preset error threshold value;
and calculating the identification accuracy according to the number of the accurate results and the total number of the identification coordinates of the current round.
According to some embodiments of the invention, the training address information includes an accurate location address, a noisy location address, and the acquiring training data set further includes:
acquiring the accurate place address and the noise place address corresponding to the accurate place address from the training address information;
And mixing the positive sample data and the negative sample data to form the training address information by taking the accurate place address as positive sample data and the noise place address as negative sample data.
According to some embodiments of the invention, the obtaining the accurate location address and the noise location address corresponding to the accurate location address from the training address information includes:
acquiring the accurate place address from the training address information, and acquiring an error address corpus matched with the accurate place address from the training address information;
and acquiring the noise place address according to the error address corpus.
According to some embodiments of the invention, the mixing the positive sample data with the negative sample data with the accurate location address as positive sample data and the noise location address as negative sample data to form the respective training address information includes:
mixing the positive sample data and the negative sample data to obtain mixed data;
adjusting the duty ratio of the positive sample data and the negative sample data in the mixed data;
And when the duty ratio of the positive sample data in the mixed data is larger than the duty ratio of the negative sample data in the mixed data, taking the adjusted mixed data as the training address information.
In a second aspect, an embodiment of the present invention provides an electronic device, including: the address information conversion device comprises a memory and a processor, wherein the memory stores a computer program, and the processor realizes the address information conversion method according to any one of the embodiments of the first aspect of the invention when executing the computer program.
In a third aspect, an embodiment of the present invention provides a computer-readable storage medium storing a program that is executed by a processor to implement the address information conversion method according to any one of the embodiments of the first aspect of the present invention.
The address information conversion method, the electronic device and the computer readable storage medium according to the embodiments of the present invention have at least the following advantages:
in the address information conversion method of the embodiment of the invention, target address information is firstly acquired, and the target address information is input into a Seq2Seq semantic recognition model, wherein the Seq2Seq semantic recognition model is obtained by training a basic recognition model through training address information and training longitude and latitude coordinates corresponding to the training address information. Further, semantic feature extraction is carried out on target address information through a Seq2Seq semantic recognition model to obtain address information features corresponding to the target address information, and then the Seq2Seq semantic recognition model is used for recognizing the address information features to obtain target longitude and latitude coordinates corresponding to the target address information. The address information conversion method of the embodiment of the invention converts the target address information into the target longitude and latitude coordinates based on the Seq2Seq semantic recognition model, does not need to rely on a network to perform online inquiry, and can realize the conversion of the address information into the corresponding longitude and latitude coordinates in an offline state.
Additional aspects and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention.
Drawings
The foregoing and/or additional aspects and advantages of the invention will become apparent and may be better understood from the following description of embodiments taken in conjunction with the accompanying drawings in which:
fig. 1 is a schematic flow chart of an address information conversion method according to an embodiment of the present invention;
FIG. 2 is a flowchart of another address information conversion method according to an embodiment of the present invention;
FIG. 3 is a flowchart illustrating another address information conversion method according to an embodiment of the present invention;
FIG. 4 is a flowchart illustrating another address information conversion method according to an embodiment of the present invention;
FIG. 5 is a flowchart illustrating another address information conversion method according to an embodiment of the present invention;
FIG. 6 is a schematic representation of a training data set provided by an embodiment of the present invention;
FIG. 7 is a flowchart illustrating another address information conversion method according to an embodiment of the present invention;
FIG. 8 is a flowchart illustrating another address information conversion method according to an embodiment of the present invention;
FIG. 9 is a flowchart illustrating another address information conversion method according to an embodiment of the present invention;
FIG. 10 is a flowchart illustrating another address information conversion method according to an embodiment of the present invention;
fig. 11 is a schematic diagram of an electronic device for executing the address information conversion method according to the embodiment of the present invention.
Detailed Description
Embodiments of the present invention are described in detail below, examples of which are illustrated in the accompanying drawings, wherein like or similar reference numerals refer to like or similar elements or elements having like or similar functions throughout. The embodiments described below by referring to the drawings are illustrative only and are not to be construed as limiting the invention.
In the description of the present invention, a number means one or more, a number means two or more, and greater than, less than, exceeding, etc. are understood to not include the present number, and above, below, within, etc. are understood to include the present number. The description of the first and second is for the purpose of distinguishing between technical features only and should not be construed as indicating or implying relative importance or implicitly indicating the number of technical features indicated or implicitly indicating the precedence of the technical features indicated.
In the description of the present invention, it should be understood that the direction or positional relationship indicated with respect to the description of the orientation, such as up, down, left, right, front, rear, etc., is based on the direction or positional relationship shown in the drawings, is merely for convenience of describing the present invention and simplifying the description, and does not indicate or imply that the apparatus or element to be referred to must have a specific orientation, be constructed and operated in a specific orientation, and thus should not be construed as limiting the present invention.
In the description of the present specification, reference to the terms "one embodiment," "some embodiments," "illustrative embodiments," "examples," "specific examples," or "some examples," etc., means that a particular feature, structure, material, or characteristic described in connection with the embodiment or example is included in at least one embodiment or example of the invention. In this specification, schematic representations of the above terms do not necessarily refer to the same embodiments or examples. Furthermore, the particular features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples.
In the description of the present invention, unless explicitly defined otherwise, terms such as arrangement, installation, connection, etc. should be construed broadly and the specific meaning of the terms in the present invention can be determined reasonably by a person skilled in the art in combination with the specific content of the technical solution. In addition, the following description of specific steps does not represent limitations on the order of steps or logic performed, and the order of steps and logic performed between steps should be understood and appreciated with reference to what is described in the embodiments.
Along with development of science and technology, in many occasions corresponding to location addresses and artificial intelligence services, address information needs to be converted into longitude and latitude, so that text semantics are converted into information needed by positioning. In the related art, the conversion of the address information into longitude and latitude is generally required, text semantics of the address information are firstly identified, then the address information is mapped into a longitude and latitude coordinate network according to administrative division such as province and city and county, and the like, and verification is carried out on the address information and the known address in an address database, so that the method is realized. However, the above method often requires accessing an address database on line, or building an extremely large address database in the local institution, so that it is difficult to convert address information into corresponding longitude and latitude in an offline situation. Therefore, how to convert address information into corresponding longitude and latitude coordinates in an offline state is a problem to be solved urgently by those skilled in the art.
The present invention aims to solve at least one of the technical problems existing in the prior art. To this end, the present invention proposes an address information conversion method, an electronic device, and a computer-readable storage medium, capable of implementing conversion of address information into corresponding longitude and latitude coordinates in an offline state.
Referring to fig. 1, an address information conversion method according to an embodiment of the first aspect of the present invention includes:
step S101, obtaining target address information, and inputting the target address information into a Seq2Seq semantic recognition model, wherein the Seq2Seq semantic recognition model is obtained by training a basic recognition model through training address information and training longitude and latitude coordinates corresponding to the training address information;
it should be noted that the Seq2Seq semantic recognition model is a variant of a recurrent neural network, and includes two parts, namely an Encoder (Encoder) and a Decoder (Decoder). The Seq2Seq model is an important model in natural language processing and can be used for machine translation, dialogue systems and automatic abstracts. The Seq2Seq semantic recognition model in some embodiments of the present invention is obtained by training the base recognition model through training address information and training longitude and latitude coordinates corresponding to the training address information. It should be understood that the purpose of training the basic recognition model by training the address information and training the longitude and latitude coordinates is to promote the capability of the basic recognition model to convert the address information into corresponding longitude and latitude coordinates in the case of acquiring the address information. When the accuracy of converting the address information into the corresponding longitude and latitude coordinates by the basic recognition model is high to a preset standard, the capability of converting the address information into the corresponding longitude and latitude coordinates by the basic recognition model is up to a level which can be practically applied, so that the basic recognition model up to the practical application level can be used as the Seq2Seq semantic recognition model for converting the target address information in the embodiment of the invention. It should be understood that the target address information refers to address information that needs to be converted, and the training address information and the training longitude and latitude coordinates corresponding to the training address information are training data that are needed in the process of training to obtain the Seq2Seq semantic recognition model.
Step S102, extracting semantic features of target address information through a Seq2Seq semantic recognition model to obtain address information features corresponding to the target address information;
according to some embodiments provided by the invention, after the Seq2Seq semantic recognition model obtains the target address information, semantic feature extraction is performed on the target address information through an encoder in the Seq2Seq semantic recognition model, so that the address information features corresponding to the target address information are obtained. According to some more specific embodiments of the present invention, the target address information is first converted into a word vector form at an embedding layer (embedding layer) of the encoder, then the target address information in the word vector form is immediately input to a hiding layer of the encoder, and address information features corresponding to the target address information are extracted by the hiding layer. It should be noted that, the hidden layer of the encoder is used for extracting the semantics of the target address information in the form of word vectors, so that an artificial intelligent network with a semantic recognition function needs to be selected as the hidden layer of the encoder. The hidden layer may have various options, including but not limited to: the method comprises the steps of taking a circulating neural network (Recurrent Neural Network, RN) as a hidden layer for extracting address information features, taking a Long Short-Term Memory (LSTM) as a hidden layer for extracting address information features, and taking a gating circulating unit structure (Gated Recurrent Unit, GRU) or other neural network models as a hidden layer for extracting address information features. It should be noted that, the common RNN has a problem of gradient guarantee or disappearance, so that the LSTM can maintain the long-term existence of the gradient by introducing the linear self-circulation unit. GRU is one of RNNs, and like LSTM, has been proposed to solve the problems of long-term memory and gradients in back propagation. Because LSTM has certain advantage in the sequence modeling problem, has long-term memory function, and is simple to realize, has solved the gradient disappearance and the gradient explosion's that exist in the long-term sequence training process problem, so in some preferred embodiments of this application, select LSTM as hidden layer, form the encoder of Seq2Seq semantic recognition model together with the embedded layer for carry out the semantic feature extraction to the target address information, obtain the address information characteristic that corresponds with the target address information. It should be appreciated that in some embodiments, the address information features exist in a vector form.
Step S103, the Seq2Seq semantic recognition model recognizes the address information characteristics to obtain the target longitude and latitude coordinates corresponding to the target address information.
According to some embodiments provided by the invention, after the sequence 2 sequence semantic recognition model extracts the address information features corresponding to the target address information, the address information features are further recognized to obtain the target longitude and latitude coordinates corresponding to the target address information. According to some more specific embodiments of the present invention, the address information features are decoded by a decoder in the Seq2Seq semantic recognition model, and since the basic recognition model is obtained by training the training address information and the training longitude and latitude coordinates corresponding to the training address information, the Seq2Seq semantic recognition model can establish a more accurate mapping relationship between the target address information and the target longitude and latitude coordinates, so that the target longitude and latitude coordinates corresponding to the target address information can be finally recognized on the basis of obtaining the target address information. The target longitude and latitude coordinates refer to longitude and latitude coordinates corresponding to the target address information, which are recognized by the Seq2Seq semantic recognition model.
In the address information conversion method of the embodiment of the invention, target address information is firstly acquired, and the target address information is input into a Seq2Seq semantic recognition model, wherein the Seq2Seq semantic recognition model is obtained by training a basic recognition model through training address information and training longitude and latitude coordinates corresponding to the training address information. Further, semantic feature extraction is carried out on target address information through a Seq2Seq semantic recognition model to obtain address information features corresponding to the target address information, and then the Seq2Seq semantic recognition model is used for recognizing the address information features to obtain target longitude and latitude coordinates corresponding to the target address information. The address information conversion method of the embodiment of the invention converts the target address information into the target longitude and latitude coordinates based on the Seq2Seq semantic recognition model, does not need to rely on a network to perform online inquiry, and can realize the conversion of the address information into the corresponding longitude and latitude coordinates in an offline state.
Referring to fig. 2, before acquiring the target address information and inputting the target address information into the Seq2Seq semantic recognition model, according to some embodiments of the present invention, further comprises:
step S201, a training data set is obtained, wherein the training data set comprises each training address information in a target region range and each training longitude and latitude coordinate corresponding to each training address information, and the target region range comprises a target place pointed by the target address information;
according to some embodiments of the present invention, training the Seq2Seq semantic recognition model is required before obtaining the target address information and inputting the target address information into the Seq2Seq semantic recognition model. In order to train the basic recognition model into the Seq2Seq semantic recognition model capable of realizing coordinate transformation, a training data set needs to be acquired in some embodiments of the invention. It should be noted that the training data set includes each training address information in a target geographical area, where the target geographical area refers to a location range of a geographical area that is divided into training targets. It should be understood that, in order for the trained Seq2Seq semantic recognition model to be able to effectively process the target address information, the target geographic area defined in the training needs to include the target location to which the target address information points. It should be emphasized that the training address information and the training longitude and latitude coordinates corresponding to the training address information are training data needed in the process of training to obtain the Seq2Seq semantic recognition model, and the part of training data is stored in the training data set. According to some exemplary embodiments of the present invention, the target region range includes a plurality of training sites, the training data set includes each training address information corresponding to each training site, and the training data set further includes each training longitude and latitude coordinate matched with each training site, where the training longitude and latitude coordinate is an accurate longitude and latitude coordinate of the training site, so that the capability of the basic recognition model to establish a mapping relationship between the training address information and the training longitude and latitude coordinate can be improved by training the basic recognition model based on the training address information and the training longitude and latitude coordinate, so that when the accuracy of the basic recognition model to convert the training address information into the corresponding training longitude and latitude coordinate is high to a preset standard, it is explained that the capability of converting the training address information into the corresponding training longitude and latitude coordinate through the basic recognition model reaches a level capable of practical application, and therefore the basic recognition model reaching a practical application level can be used as the Seq2Seq semantic recognition model for converting the target address information in the embodiment of the present invention.
Step S202, optimizing and training the basic recognition model based on the training data set to obtain a Seq2Seq semantic recognition model.
According to some embodiments provided by the invention, optimizing training refers to training that optimizes the ability of a base recognition model to establish a mapping relationship between training address information and training longitude and latitude coordinates. The link of extracting the address information features is carried out in a basic encoder of a basic recognition model, and the link of recognizing the address information features and obtaining the target longitude and latitude coordinates corresponding to the target address information is carried out in a basic decoder of the basic recognition model. According to some embodiments provided by the application, each time a group of training address information and corresponding training longitude and latitude coordinates are input to the basic recognition model for optimization training, the capability of the basic recognition model for mapping from the training address information to the training longitude and latitude coordinates can be improved once, therefore, a plurality of groups of training address information and corresponding training longitude and latitude coordinates are input to the basic recognition model for optimization training, and the capability of the basic recognition model for mapping from each training address information to the corresponding training longitude and latitude coordinates can be improved.
Referring to fig. 3, according to some embodiments of the present invention, the base recognition model is optimally trained based on a training dataset to obtain a Seq2Seq semantic recognition model, comprising:
step S301, performing iterative training on the basic recognition model based on a training data set, and inputting each training address information into the basic recognition model for performing the own training process in each round of iterative training to obtain each own-round recognition coordinate corresponding to each training address information;
it should be noted that, the purpose of iterative training is to gradually improve the capability of the basic identification model to map from each training address information to the corresponding training longitude and latitude coordinates after training for obtaining training longitude and latitude coordinates through several rounds of training according to the training address information. According to some embodiments provided herein, the process of obtaining training longitude and latitude coordinates by training the basic recognition model according to training address information includes: and inputting the training data set into the basic recognition model for iterative training, and after each round of iterative training, calculating the recognition accuracy of the basic recognition model and updating the basic recognition model. The basic recognition model can be obtained by pre-training a preset model which does not have the capability of obtaining the training longitude and latitude coordinates according to the training address information recognition, or can be obtained by selecting a preset model which has the capability of obtaining the training longitude and latitude coordinates according to the training address information recognition. According to some exemplary embodiments of the present invention, in each iteration training, each training address information is input into a basic recognition model to perform a local training process, and each local recognition coordinate corresponding to each training address information is obtained. According to some more preferred embodiments of the invention, each training address information needs to undergo a plurality of rounds of iterative training in the process of optimizing training until the basic recognition model can map each training address information to the corresponding training longitude and latitude coordinates with the recognition accuracy reaching the preset standard, and then the capability of the basic recognition model from the training address information to the training longitude and latitude coordinates is improved once.
Step S302, after each round of iterative training, comparing each round of identification coordinates with corresponding training longitude and latitude coordinates, calculating the identification accuracy of a basic identification model and updating parameters of the basic identification model;
according to some embodiments provided by the invention, the present round of recognition coordinates refer to coordinates obtained by recognizing training address information through a basic recognition model in the present round of iterative training. It should be noted that the purpose of iterative training is to continuously optimize the capability of the basic recognition model to map from each training address information to the corresponding training longitude and latitude coordinates. After each round of iterative training, the recognition accuracy of the basic recognition model needs to be calculated, so that the capability of mapping the basic recognition model from each training address information to the corresponding training longitude and latitude coordinates in the training process is improved gradually. It should be noted that, in the process of optimizing training, each time an iteration training is performed, the basic recognition model needs to be updated once, so that the basic recognition model can be mapped from each training address information to each corresponding training longitude and latitude coordinate in the next iteration training, and better performance can be achieved. In iterative training, updating the basic recognition model is mainly completed by adjusting parameters of the basic recognition model, namely, internal parameters of the basic recognition model, which are related to the capability of mapping each training address information to the corresponding training longitude and latitude coordinates. According to some exemplary embodiments of the present invention, the recognition accuracy of the basic recognition model is calculated by comparing each of the recognition coordinates of the present wheel with the corresponding training longitude and latitude coordinates. It should be understood that the recognition coordinates of the present round refer to coordinates obtained by recognizing the training address information through the basic recognition model in the iterative training of the present round, and the training longitude and latitude coordinates are accurate coordinates matched with the training address information, so that the recognition accuracy of the basic recognition model can be obtained by comparing the recognition coordinates of the present round with the training longitude and latitude coordinates.
Step S303, counting the change condition of the recognition accuracy after each round of iterative training, and obtaining a trained Seq2Seq semantic recognition model when the recognition accuracy is converged to a first fixed value.
It should be noted that, when the recognition accuracy rate converges to a first fixed value, the iterative training can be stopped and a trained Seq2Seq semantic recognition model is obtained, where the first fixed value refers to: after the internal parameters of the basic recognition model are adjusted in a plurality of rounds, the recognition accuracy rate of the basic recognition model converges. In some embodiments provided herein, the recognition accuracy of the basic recognition model will be stabilized within a certain error interval, for example, the recognition accuracy of the basic recognition model fluctuates within a range of 86% to 88%, and the fixed value may be considered to be 87%. It will be appreciated that the first constant value is not an exact constant value, but a value that varies with the training conditions. When the recognition accuracy of the basic recognition model is converged to a first fixed value, judging that the optimization training has achieved a better effect, and stopping the iterative training, wherein the basic recognition model obtained after stopping the iterative training, namely the Seq2Seq semantic recognition model, is subjected to optimization training, and the capability of mapping each training address information to the corresponding training longitude and latitude coordinates is improved. According to some embodiments provided herein, as iterative training proceeds, the capability of the basic recognition model to map each training address information to its corresponding training longitude and latitude coordinates is improved, and the step of extracting the address information features is performed in the basic encoder of the basic recognition model, and the step of recognizing the address information features and obtaining the destination longitude and latitude coordinates corresponding to the target address information is performed in the basic decoder of the basic recognition model. Therefore, the underlying reason that the basic recognition model is optimized is that the capability of the basic encoder to extract the address information features is optimized, and the capability of the basic decoder to extract the semantic features of the target address information and obtain the address information features corresponding to the target address information is optimized.
Referring to fig. 4, according to some embodiments of the present invention, the basic recognition model is iteratively trained based on a training data set, and in each iteration training, each training address information is input into the basic recognition model to perform a present training process, so as to obtain each present-round recognition coordinate corresponding to each training address information, including:
step S401, in each iteration training, randomly starting semantic recognition neurons in a batch of basic recognition models;
step S402, based on the randomly opened semantic recognition neurons, the training address information is subjected to the training processing of the round, and the recognition coordinates of the round are obtained.
It should be noted that, the purpose of iterative training is to gradually improve the capability of the basic identification model to map from each training address information to the corresponding training longitude and latitude coordinates after training for obtaining training longitude and latitude coordinates through several rounds of training according to the training address information. In a machine-learned model, if the parameters of the model are too many and the training samples are too few, the trained model is easy to generate the phenomenon of over-fitting. The problem of overfitting is often encountered when training neural networks, the overfitting specific body performs: the model has smaller loss function on training data and higher prediction accuracy; however, the loss function is larger on the test data, and the prediction accuracy is lower. Overfitting is a common problem of many machine studies, and in order to avoid overfitting during the process of optimally training the underlying recognition model, some preferred embodiments of the present invention improve the performance of the neural network by preventing the feature detectors from acting together. Specifically, in each round of iterative training, a group of semantic recognition neurons in the basic recognition model are randomly opened, and then each round of training processing is performed on each training address information based on the randomly opened semantic recognition neurons, so that each round of recognition coordinates are obtained. It should be understood that, in each iteration training, although a part of semantic recognition neurons in the basic recognition model are randomly opened to participate in training, many semantic recognition neurons arranged in the basic recognition model can still be sufficiently trained after a plurality of iteration training. And because the semantic recognition neurons used in each round of iterative training are randomly selected to be opened, each semantic recognition neuron arranged in the basic recognition model can have a larger weight influence on recognition of training address information after a plurality of rounds of iterative training. Therefore, under the condition that the number of the semantic recognition neurons is randomly opened, the basic recognition model is subjected to iterative training for several rounds to obtain the Seq2Seq semantic recognition model, and the overfitting phenomenon in the Seq2Seq semantic recognition model can be well relieved. It should be understood that, in each iteration training, each training address information is input into the basic recognition model to perform the present training process, and obtaining each present recognition coordinate corresponding to each training address information may include, but is not limited to, the above embodiment.
According to some embodiments of the present invention, the objective of iterative training is to continuously optimize the ability of the underlying recognition model to map from each training address information to the respective corresponding training longitude and latitude coordinates. After each round of iterative training, the recognition accuracy of the basic recognition model needs to be calculated, so that the capability of mapping the basic recognition model from each training address information to the corresponding training longitude and latitude coordinates in the training process is improved gradually. According to some exemplary embodiments of the present invention, the recognition accuracy of the basic recognition model is calculated by comparing each of the recognition coordinates of the present wheel with the corresponding training longitude and latitude coordinates. It should be understood that the recognition coordinates of the present round refer to coordinates obtained by recognizing the training address information through the basic recognition model in the iterative training of the present round, and the training longitude and latitude coordinates are accurate coordinates matched with the training address information, so that the recognition accuracy of the basic recognition model can be obtained by comparing the recognition coordinates of the present round with the training longitude and latitude coordinates.
According to some more specific embodiments of the present invention, in order to calculate the recognition accuracy of the basic recognition model, it is necessary to obtain each coordinate interval error according to each principal moment recognition coordinate and each training longitude and latitude coordinate, and then obtain the accurate result number that the coordinate interval error is smaller than the preset error threshold value, and further calculate the recognition accuracy according to the accurate result number and the total number of the principal moment recognition coordinates.
Referring to fig. 5, after each iteration of training, each current round of recognition coordinates is compared with corresponding training longitude and latitude coordinates, and recognition accuracy of a basic recognition model is calculated and parameters of the basic recognition model are updated, including:
step S501, according to the identification coordinates of each principal wheel and each training longitude and latitude coordinate, obtaining each coordinate interval error;
according to some exemplary embodiments of the present invention, in order to calculate the recognition accuracy of the basic recognition model, each of the recognition coordinates of the present wheel needs to be compared with the corresponding training longitude and latitude coordinates. In the process of comparing each own-round identification coordinate with the corresponding training longitude and latitude coordinate, the own-round identification coordinate is obtained by identifying training address information through a basic identification model in the own-round iterative training; the training longitude and latitude coordinates are accurate coordinates matched with training address information. It should be understood that the recognition coordinates of the present wheel and the training longitude and latitude coordinates to be compared need to correspond to the same training address information.
According to some more specific embodiments of the present invention, referring to the training data set schematic chart shown in fig. 6, if the training is optimized based on the training address information a, after each iteration training, the present round of recognition coordinates and the training longitude and latitude coordinates "120, which are obtained by recognizing the recognition training address information a, will be recognized; 20", and obtaining a coordinate interval error delta a; similarly, if the optimization training is performed based on the training address information B, after each round of iterative training, the local round of recognition coordinates and the training longitude and latitude coordinates obtained by recognizing the recognition training address information B are 800;96", to obtain a coordinate interval error delta b; if the optimization training is performed based on the training address information C, after each round of iterative training, the round of identification coordinates and training longitude and latitude coordinates 90 which are obtained by identifying the training address information C are identified; 45", and obtaining a coordinate interval error delta c; if the optimization training is performed based on the training address information D, after each round of iterative training, the round of recognition coordinates and training longitude and latitude coordinates obtained by recognizing the recognition training address information D are obtained 150; 23' to obtain a coordinate interval error delta d; if the optimization training is performed based on the training address information E, after each round of iterative training, the current round of identification coordinates and training longitude and latitude coordinates 112 obtained by identifying the training address information E are identified; 800", and obtaining a coordinate interval error delta e; if the optimization training is performed based on the training address information F, after each round of iterative training, the current round of identification coordinates and training longitude and latitude coordinates' 30 are obtained by identifying the training address information F; 60", to obtain the coordinate interval error deltaf.
Step S502, obtaining accurate result quantity that the coordinate interval error is smaller than a preset error threshold value;
step S503, calculating the recognition accuracy according to the number of the accurate results and the total number of the recognition coordinates of the round.
According to some embodiments of the invention, the preset error threshold refers to a preset error threshold. When the coordinate interval error is greater than or equal to a preset error threshold value, the difference between the current wheel identification coordinate and the training longitude and latitude coordinate is too large, and the current wheel identification coordinate cannot be regarded as an accurate result of current wheel identification of the basic identification model; when the coordinate interval error is smaller than the preset error threshold, the difference between the current wheel identification coordinate and the training longitude and latitude coordinate is smaller, and the current wheel identification coordinate can be regarded as an accurate result of current wheel identification of the basic identification model. It should be noted that, the purpose of obtaining the number of accurate results with the coordinate interval error smaller than the preset error threshold is to count the accurate results in the iterative training of the present round, so as to further calculate the recognition accuracy according to the number of accurate results and the total number of the recognition coordinates of the present round. It should be understood that calculating the recognition accuracy can be converted to a percentage by taking the number of accurate results as a numerator and the total number of recognition coordinates of the present round as a denominator. It should be appreciated that calculating the recognition accuracy of the underlying recognition model may include, but is not limited to, the embodiments described above.
According to some embodiments of the present invention, when the address information conversion method is actually used, various types of address information similar to the target address information are often encountered. For example, when the target address information is "XX literature teaching road", similar address information may include similar address information of "XX county literature teaching road", "XX literature teaching road primary school", "XX literature teaching two road" and so on. Therefore, in order to improve the conversion accuracy of the address information conversion method of the present invention, in some exemplary embodiments of the present invention, obtaining the training data set of the basic recognition model further includes obtaining an accurate location address and a noise location address corresponding to the accurate location address from each piece of training address information, and further mixing the positive sample data and the negative sample data to form each piece of training address information with the accurate location address as positive sample data and the noise location address as negative sample data.
Referring to fig. 7, according to some embodiments of the present invention, each training address information includes an accurate location address, a noisy location address, and acquiring a training data set, further including:
step S701, obtaining accurate place addresses and noise place addresses corresponding to the accurate place addresses from the training address information;
In step S702, the accurate location address is used as positive sample data, the noise location address is used as negative sample data, and the positive sample data and the negative sample data are mixed to form each training address information.
According to some embodiments provided herein, the accurate location address is an address pointing directly to the training location, and the noisy location address refers to a location address that approximates the accurate location address. It is understood that the basic recognition model is optimized and trained based on the accurate place address, so that the recognition accuracy of the basic recognition model is continuously improved, and the capability of mapping the basic recognition model from each training address information to each corresponding training longitude and latitude coordinate can be continuously optimized; the accurate place address and the noise place address are mixed to form training address information, then the training address information after mixing is used for carrying out optimization training on the basic recognition model, the recognition accuracy of the basic recognition model is continuously improved, the robustness of the basic recognition model can be optimized while the capability of the basic recognition model for mapping from each training address information to the corresponding training longitude and latitude coordinates is continuously optimized, and the anti-interference capability of the basic recognition model in the recognition conversion process is improved. It should be appreciated that acquiring the training data set may include, but is not limited to, the embodiments described above.
Referring to fig. 8, according to some embodiments of the present invention, acquiring an accurate location address, a noise location address corresponding to the accurate location address, from each training address information includes:
step S801, obtaining accurate location addresses from each piece of training address information, and obtaining error address text sets matched with the accurate location addresses from each piece of training address information;
step S802, according to the error address corpus, a noise location address is obtained.
According to some exemplary embodiments provided herein, the more similar the noisy location address is to the accurate location address, the more disturbing the accurate location address. Therefore, in order to continuously optimize the capability of the basic recognition model for mapping from each training address information to each corresponding training longitude and latitude coordinate, the robustness of the basic recognition model is optimized, and the anti-interference capability of the basic recognition model in the recognition conversion process is improved. In some preferred embodiments of the present invention, an accurate location address is obtained from each training address information, and an error address corpus matching the accurate location address is obtained from each training address information, so as to obtain a noise location address according to the error address corpus. The error address corpus is a corpus composed of location addresses that are similar to the exact location addresses, and the location addresses in the file may be arranged according to, but not limited to, the approximation of the exact location addresses. It should be appreciated that obtaining the accurate location address, the noisy location address corresponding to the accurate location address, from the respective training address information may include, but is not limited to, the embodiments described above.
Referring to fig. 9, according to some embodiments of the present invention, mixing positive sample data with negative sample data with accurate location address as positive sample data and noise location address as negative sample data to form respective training address information includes:
step S901, mixing positive sample data and negative sample data to obtain mixed data;
according to some exemplary embodiments of the present invention, the accurate venue address is an address that points directly to the training venue; and the exact location address is an address that points directly to the training location. Therefore, in order to continuously optimize the capability of the basic recognition model for mapping from each training address information to each corresponding training longitude and latitude coordinate, the robustness of the basic recognition model is optimized, and the anti-interference capability of the basic recognition model in the recognition conversion process is improved. Positive sample data in training address information is sampled from an accurate place address, negative sample data in training address information is sampled from a noise place address, the training address information formed by mixing in the mode is used for optimizing training, when the recognition accuracy of a basic recognition model rises and converges to a first fixed value, the end of optimizing training can be judged, and a Seq2Seq semantic recognition model after optimizing training is completed is obtained. It is understood that the Seq2Seq semantic recognition model obtained through the optimization training can be mapped to the corresponding training longitude and latitude coordinates from each training address information, and has better anti-interference capability in the recognition conversion process.
Step S902, the duty ratio of the positive sample data and the negative sample data in the mixed data is adjusted;
in step S903, when the duty ratio of the positive sample data in the mixed data is greater than the duty ratio of the negative sample data in the mixed data, the adjusted mixed data is used as each training address information.
According to some embodiments provided by the invention, if the positive sample data in the mixed data is too small in proportion, the recognition accuracy of the basic recognition model is not conducive to being guided to be improved in the optimization training process. Thus, the positive sample data is required to be a majority of the mixed data, and in some embodiments of the present invention, the positive sample data is required to be adjusted to the majority of the mixed data before training address information is generated from the mixed data. It should be clear that the ratio of positive sample data in the mixed data refers to the ratio of positive sample data in the total mixed data, and the ratio of negative sample data in the mixed data refers to the ratio of positive sample data in the total mixed data. It should be noted that, when the duty ratio of the positive sample data in the mixed data is greater than the duty ratio of the negative sample data in the mixed data, it is explained that the accurate place address in the positive sample data occupies the main part in the mixed data, and the training address information obtained by using this situation is used for optimization training, which is beneficial to guiding the recognition accuracy of the basic recognition model to be improved.
Referring to fig. 10, according to some embodiments of the present invention, the base recognition model is optimally trained based on the training data set to obtain a Seq2Seq semantic recognition model, further comprising:
step S1001, after each round of iterative training, comparing each round of identification coordinates with corresponding training longitude and latitude coordinates, calculating a loss function output value of a basic identification model, and updating parameters of the basic identification model;
according to some embodiments provided by the invention, the present round of recognition coordinates refer to coordinates obtained by recognizing training address information through a basic recognition model in the present round of iterative training. It should be noted that the purpose of iterative training is to continuously optimize the capability of the basic recognition model to map from each training address information to the corresponding training longitude and latitude coordinates. After each round of iterative training, the output value of the loss function of the basic recognition model needs to be calculated so as to clearly optimize the capability of the basic recognition model for mapping from each training address information to the corresponding training longitude and latitude coordinates in the training process. It should be noted that, in the process of optimizing training, each time an iteration training is performed, the basic recognition model needs to be updated once, so that the basic recognition model can be mapped from each training address information to each corresponding training longitude and latitude coordinate in the next iteration training, and better performance can be achieved. In iterative training, updating the basic recognition model is mainly completed by adjusting parameters of the basic recognition model, namely, internal parameters of the basic recognition model, which are related to the capability of mapping each training address information to the corresponding training longitude and latitude coordinates.
Step S1002, counting the change condition of the output value of the loss function after each round of iterative training, and obtaining a trained Seq2Seq semantic recognition model when the output value of the loss function is converged to a second fixed value.
It should be noted that, when the output value of the loss function converges to the second constant value, the iterative training may be stopped and a trained Seq2Seq semantic recognition model may be obtained, where the second constant value refers to: after the internal parameters of the basic recognition model are adjusted in a plurality of rounds, the loss function of the basic recognition model outputs a convergence value. In some embodiments provided herein, the output value of the loss function of the basic identification model will be stabilized within a certain error interval, for example, the output value of the loss function of the basic identification model fluctuates within a range of 6% to 8%, and the fixed value can be considered to be 7%. It will be appreciated that the second constant is not an exact constant value, but a value that varies with the training conditions. When the loss function output value of the basic recognition model is converged to a second fixed value, judging that the optimization training has achieved a better effect, and stopping the iterative training, wherein the basic recognition model obtained after stopping the iterative training, namely the Seq2Seq semantic recognition model, is subjected to optimization training, and the capability of mapping each training address information to the corresponding training longitude and latitude coordinates is improved.
Fig. 11 shows an electronic device 1100 provided by an embodiment of the invention. The electronic device 1100 includes: a processor 1101, a memory 1102, and a computer program stored on the memory 1102 and executable on the processor 1101, the computer program when run is for performing the address information conversion method described above.
The processor 1101 and the memory 1102 may be connected by a bus or other means.
The memory 1102 is used as a non-transitory computer readable storage medium for storing non-transitory software programs and non-transitory computer executable programs, such as address information translation methods described in embodiments of the present invention. The processor 1101 implements the address information conversion method described above by running non-transitory software programs and instructions stored in the memory 1102.
Memory 1102 may include a storage program area that may store an operating system, at least one application program required for functionality, and a storage data area. The storage data area may store address information conversion methods described above. In addition, the memory 1102 may include high-speed random access memory 1102, and may also include non-transitory memory 1102, such as at least one storage device memory device, flash memory device, or other non-transitory solid state memory device. In some implementations, the memory 1102 optionally includes memory 1102 remotely located relative to the processor 1101, the remote memory 1102 being connectable to the electronic device 1100 through a network. Examples of such networks include, but are not limited to, the internet, intranets, local area networks, mobile communication networks, and combinations thereof.
The non-transitory software programs and instructions required to implement the address information conversion method described above are stored in the memory 1102 and when executed by the one or more processors 1101, perform the address information conversion method described above, for example, perform method steps S101 through S103 in fig. 1, method steps S201 through S202 in fig. 2, method steps S301 through S303 in fig. 3, method steps S401 through S402 in fig. 4, method steps S501 through S503 in fig. 5, method steps S701 through S702 in fig. 7, method steps S801 through S802 in fig. 8, method steps S901 through S903 in fig. 9, and method steps S1001 through S1002 in fig. 10.
The embodiment of the invention also provides a computer readable storage medium which stores computer executable instructions for executing the address information conversion method.
In an embodiment, the computer-readable storage medium stores computer-executable instructions that are executed by one or more control processors, for example, to perform method steps S101 through S103 in fig. 1, method steps S201 through S202 in fig. 2, method steps S301 through S303 in fig. 3, method steps S401 through S402 in fig. 4, method steps S501 through S503 in fig. 5, method steps S701 through S702 in fig. 7, method steps S801 through S802 in fig. 8, method steps S901 through S903 in fig. 9, and method steps S1001 through S1002 in fig. 10. The above described apparatus embodiments are merely illustrative, wherein the units illustrated as separate components may or may not be physically separate, i.e. may be located in one place, or may be distributed over a plurality of network elements. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.
Those of ordinary skill in the art will appreciate that all or some of the steps, systems, and methods disclosed above may be implemented as software, firmware, hardware, and suitable combinations thereof. Some or all of the physical components may be implemented as software executed by a processor, such as a central processing unit, digital signal processor, or microprocessor, or as hardware, or as an integrated circuit, such as an application specific integrated circuit. Such software may be distributed on computer readable media, which may include computer storage media (or non-transitory media) and communication media (or transitory media). The term computer storage media includes both volatile and nonvolatile, removable and non-removable media implemented in any method or technology for storage of information such as computer readable instructions, data structures, program modules or other data, as known to those skilled in the art. Computer storage media includes, but is not limited to, RAM, ROM, EEPROM, flash memory or other memory technology, CD-ROM, digital Versatile Disks (DVD) or other optical disk storage, magnetic cassettes, magnetic tape, storage device storage or other magnetic storage devices, or any other medium which can be used to store the desired information and which can be accessed by a computer. Furthermore, as is well known to those of ordinary skill in the art, communication media typically include computer readable instructions, data structures, program modules, or other data in a modulated data signal, such as a carrier wave or other transport mechanism, and may include any information delivery media. It should also be appreciated that the various embodiments provided by the embodiments of the present invention may be arbitrarily combined to achieve different technical effects.
While the preferred embodiment of the present invention has been described in detail, the present invention is not limited to the above embodiments, and those skilled in the art can make various equivalent modifications or substitutions without departing from the spirit and scope of the present invention, and these equivalent modifications or substitutions are included in the scope of the present invention as defined in the appended claims.
Claims (7)
1. A method of address information conversion, the method comprising:
acquiring target address information, inputting the target address information into a Seq2Seq semantic recognition model, wherein the Seq2Seq semantic recognition model is obtained by training a basic recognition model through training address information and training longitude and latitude coordinates corresponding to the training address information; the training of the basic recognition model to obtain the Seq2Seq semantic recognition model comprises the following steps:
acquiring a training data set, wherein the training data set comprises training address information and training longitude and latitude coordinates corresponding to the training address information in a target region range, and the target region range comprises a target place pointed by the target address information;
performing iterative training on the basic recognition model based on the training data set, inputting each training address information into the basic recognition model in each round of iterative training, randomly starting a batch of semantic recognition neurons in the basic recognition model, performing the training treatment on each training address information round based on the semantic recognition neurons which are randomly started, and obtaining each round of recognition coordinates corresponding to each training address information round;
After each round of iterative training, comparing the identification coordinates of each round with the longitude and latitude coordinates of each training, calculating the identification accuracy of the basic identification model and updating the parameters of the basic identification model;
counting the change condition of the recognition accuracy after each round of iterative training, and obtaining a trained Seq2Seq semantic recognition model when the recognition accuracy is converged to a first certain value;
and extracting semantic features of the target address information through the trained Seq2Seq semantic recognition model to obtain address information features corresponding to the target address information, and recognizing the address information features to obtain target longitude and latitude coordinates corresponding to the target address information.
2. The method according to claim 1, wherein after each iteration of training, comparing the recognition coordinates of each present round with the longitude and latitude coordinates of each training, calculating the recognition accuracy of the basic recognition model and updating the parameters of the basic recognition model, including:
obtaining each coordinate interval error according to the identification coordinates of each own wheel and the corresponding training longitude and latitude coordinates;
Acquiring the number of accurate results of which the coordinate interval error is smaller than a preset error threshold value;
and calculating the identification accuracy according to the number of the accurate results and the total number of the identification coordinates of the current round.
3. The method of claim 1 or 2, wherein each training address information includes an accurate location address, a noisy location address, the acquiring training data set, further comprising:
acquiring the accurate place address and the noise place address corresponding to the accurate place address from the training address information;
the accurate place address is used as positive sample data, the noise place address is used as negative sample data, and the positive sample data and the negative sample data are mixed to form the training data set.
4. A method according to claim 3, wherein said obtaining the accurate location address, the noisy location address corresponding to the accurate location address, from the respective training address information comprises:
acquiring the accurate place address from the training address information, and acquiring an error address corpus matched with the accurate place address from the training address information;
And acquiring the noise place address according to the error address corpus.
5. A method according to claim 3, wherein said mixing said positive sample data with said negative sample data with said accurate location address as positive sample data and said noisy location address as negative sample data, comprises:
mixing the positive sample data and the negative sample data to obtain mixed data;
adjusting the duty ratio of the positive sample data and the negative sample data in the mixed data;
and when the duty ratio of the positive sample data in the mixed data is larger than the duty ratio of the negative sample data in the mixed data, taking the adjusted mixed data as the training data set.
6. An electronic device, comprising: a memory, a processor storing a computer program, the processor implementing the address information conversion method according to any one of claims 1 to 5 when executing the computer program.
7. A computer-readable storage medium storing a program that is executed by a processor to implement the address information conversion method according to any one of claims 1 to 5.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210908906.8A CN115277626B (en) | 2022-07-29 | 2022-07-29 | Address information conversion method, electronic device, and computer-readable storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210908906.8A CN115277626B (en) | 2022-07-29 | 2022-07-29 | Address information conversion method, electronic device, and computer-readable storage medium |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN115277626A CN115277626A (en) | 2022-11-01 |
| CN115277626B true CN115277626B (en) | 2023-07-25 |
Family
ID=83746380
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210908906.8A Active CN115277626B (en) | 2022-07-29 | 2022-07-29 | Address information conversion method, electronic device, and computer-readable storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115277626B (en) |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2020140377A1 (en) * | 2019-01-04 | 2020-07-09 | 平安科技（深圳）有限公司 | Neural network model training method and apparatus, computer device, and storage medium |
Family Cites Families (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108319585B (en) * | 2018-01-29 | 2021-03-02 | 北京三快在线科技有限公司 | Data processing method and device, electronic equipment and computer readable medium |
| CN109960795B (en) * | 2019-02-18 | 2024-05-07 | 平安科技（深圳）有限公司 | Address information standardization method and device, computer equipment and storage medium |
| JP7041281B2 (en) * | 2019-07-04 | 2022-03-23 | 浙江大学 | Address information feature extraction method based on deep neural network model |
| US11462208B2 (en) * | 2019-09-11 | 2022-10-04 | Oracle International Corporation | Implementing a correction model to reduce propagation of automatic speech recognition errors |
| CN111639163A (en) * | 2020-04-29 | 2020-09-08 | 深圳壹账通智能科技有限公司 | Problem generation model training method, problem generation method and related equipment |
| CN113792107A (en) * | 2021-09-15 | 2021-12-14 | 北京沃东天骏信息技术有限公司 | Region identification method and device |
-
2022
- 2022-07-29 CN CN202210908906.8A patent/CN115277626B/en active Active
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2020140377A1 (en) * | 2019-01-04 | 2020-07-09 | 平安科技（深圳）有限公司 | Neural network model training method and apparatus, computer device, and storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN115277626A (en) | 2022-11-01 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US11488067B2 (en) | Training machine learning models using teacher annealing | |
| CN115080764B (en) | Medical similar entity classification method and system based on knowledge graph and clustering algorithm | |
| CN113220839B (en) | Intention identification method, electronic equipment and computer readable storage medium | |
| CN110462638B (en) | Training neural networks using posterior sharpening | |
| EP3620994A1 (en) | Methods, apparatuses, devices, and computer-readable storage media for determining category of entity | |
| CN111291099B (en) | Address fuzzy matching method and system and computer equipment | |
| CN113065525B (en) | Age identification model training method, face age identification method and related device | |
| CN110929640B (en) | Wide remote sensing description generation method based on target detection | |
| CN113592037A (en) | Address matching method based on natural language inference | |
| Song et al. | Hidden Markov model and driver path preference for floating car trajectory map matching | |
| CN111161238A (en) | Image quality evaluation method and device, electronic device, and storage medium | |
| CN111241106A (en) | Approximate data processing method, device, medium and electronic equipment | |
| CN115277626B (en) | Address information conversion method, electronic device, and computer-readable storage medium | |
| CN114926701A (en) | Model training method, target detection method and related equipment | |
| CN117575008A (en) | Training sample generation method, model training method, knowledge question-answering method and knowledge question-answering device | |
| Wang et al. | Gated recurrent unit with RSSIs from heterogeneous network for mobile positioning | |
| CN117034031A (en) | Sentence generation method and device based on communication scene | |
| CN112528062A (en) | Cross-modal weapon retrieval method and system | |
| CN116127343A (en) | Household transformer relation and phase identification method and system based on voltage sequence similarity | |
| CN110287487A (en) | The recognition methods of subject-predicate language, device, equipment and computer readable storage medium | |
| CN113420680B (en) | Remote sensing image area attention and text generation method based on GRU attention | |
| CN112990292B (en) | Method and device for generating dialogue state based on neural network | |
| CN112149005B (en) | Method, apparatus, device and readable storage medium for determining search results | |
| CN113723712A (en) | Wind power prediction method, system, device and medium | |
| CN114416941A (en) | Generation method and device of dialogue knowledge point determination model fusing knowledge graph |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |